Ganesh Datta
HostCTO & Co-founder of Cortex
Martin Nettling
Senior Director of Engineering
July 16, 2026
In This Episode
Martin Nettling, Ph.D., is a Senior Director of Engineering at Tealium, where he also runs QA. He came up through bioinformatics and molecular genetics before moving into computer science and software development. Martin was VP of Engineering at Datameer before joining Tealium two years ago. He brought six or seven engineers with him, which meant grafting an established team and its own way of working onto a worldwide company with teams in Poland and the US. That experience of building trust across cultures runs straight into the topic of this episode.
Martin joins Cortex CTO Ganesh Datta to separate two ideas most teams lump together. Confidence is for tools, trust is for people, and treating an AI-generated PR as a referendum on the engineer who wrote it gets both wrong. They get into the "augmented author model" (whoever uses AI owns how its output is communicated), the "unbroken why" that lets you trace a log line back to company strategy, why throwing every kind of code into one review bucket is a mistake, and how risk labels bubbling up into a traffic-light system let teams stop reviewing everything. Martin closes on the part of the system he trusts least, and it isn't the model. Read more from Dr. Nettling on trust and confidence here.
You’ll learn
Confidence is for tools, trust is for people, and mixing them up wrecks code review. When Martin came down hard on an engineer for an AI-generated pull request, he realized he'd translated low confidence in the tool into mistrust of the person. Separating the two changed the conversation from "this doesn't work" to "show me how you used it, can we improve it?"
The augmented author model: if you used AI, you own how it's communicated. One sentence of intent attached to the artifact is enough to start rebuilding trust, and once people expect it from each other, it becomes culture.
An "unbroken why" earns the right to stop reviewing. Trust comes from being able to trace a production log entry back through the ticket to the company strategy, and back down again. When that chain holds and proves itself, you can stop inspecting individual links.
Stop throwing every kind of code into one bucket. A calorie-tracking app you'll delete in a year and code that has to live in a bank for twenty years deserve different scrutiny. The review should match the stakes.
The least trustworthy part of the system? The human. People are usefully lazy, which is why we automate, but they occasionally skip the steps that matter.
Quotes
You have confidence in a tool, but you have trust in a human person.
Martin Nettling
Senior Director of Engineering
You use AI, and you take this interaction with the tool and you are now responsible for how it's transported and how it's communicated. That’s the gap right now. So often people do something with AI and just hand it over without any comment.
Martin Nettling
Senior Director of Engineering
Some people say 'I'm doing a code review.' No. You're making sure the codebase is only filled with high-quality, trusted code. The review is just the tool that starts it
Martin Nettling
Senior Director of Engineering
Humans can misuse anything... We do not always interpret the results of AI as critically as we should.
Martin Nettling
Senior Director of Engineering
Timestamps
(01:11)
Martin's background, from bioinformatics to VP of Engineering, and bringing his team into Tealium.
(05:00)
Trust vs. confidence, explained through an Uber-driver analogy.
(06:48)
The AI-generated pull request that reframed how he gives feedback.
(09:36)
The augmented author model: owning how AI output is communicated.
(13:39)
Why you can't throw every kind of code into one review bucket.
(16:25)
The "unbroken why" and tracing a log line back to strategy.
(20:38)
Reasoning bias and AI litter vs. AI slop.
(24:06)
What code review is actually for.
(26:56)
What can be automated with confidence, and the load test an agent couldn't run.
(31:40)
Ownership gaps, observability, and repo hygiene as trust signals.
(34:15)
Risk labels on PRs and a green/yellow/red system that bubbles up.
(39:53)
The least trustworthy part of the system: the human.
Other episodes
July 9, 2026Your Ops Review is Theater, and That's the Point: Aleks Rudzitis on Turning Reliability into a Shared Value
Aleks Rudzitis
Principal Engineer
July 9, 2026Operational Excellence (OpEx) Reviews: The Weekly Meeting That Actually Changes Behavior
Shawn Burke
Distinguished Engineer
July 9, 2026Okta's Dinesh Sukhija on Meeting AI with AI, and the Convergence of Platform, SRE, and Security
Dinesh Sukhija
Director of Engineering
